Coin and banknote acceptors are well known. One example of a coin acceptor is described in our GB-A-2 169 429. The acceptor includes a coin rundown path along which coins pass through a coin sensing station at which sensor coils perform a series of inductive tests on the coins in order to develop coin parameter signals which are indicative of the material and metallic content of the coin under test. The coin parameter signals are digitised and compared with stored coin data by means of a microcontroller to determine the acceptability or otherwise of the test coin. If the coin is found to be acceptable, the microcontroller operates an accept gate so that the coin is directed to an accept path. Otherwise, the accept gate remains inoperative and the coin is directed to a reject path.
In banknote validators, sensors detect characteristics of the banknote. For example, optical detectors can be used to detect the geometrical size of the banknote, its spectral response to a light source in transmission or reflection, or the presence of magnetic printing ink can be detected with an appropriate sensor. The parameter signals thus developed are digitised and compared with stored values in a similar way to the previously described prior art coin acceptor. The acceptability of the banknote is determined on the basis of the results of the comparison.
When a number of coins or banknotes of the same denomination are passed through an acceptor, successive values of coin or banknote parameter data are thus developed. When the distribution of the values of these signals is plotted as a graph, the result is a bell curve, with a central peak and tails on opposite sides. The shape of the graph may typically although not necessarily be Gaussian.
The distribution illustrates that for a money item, such as a coin or banknote of a particular denomination, the most probable value of the corresponding parameter signal lies at the peak of the bell curve, with a decreasing probability to either side. In prior coin and banknote acceptors data is stored in a memory, corresponding to acceptable ranges of parameter signal for a particular denomination. The acceptor compares the value for a coin or banknote under test with the stored data to determine authenticity. The data may define windows in terms of upper and lower limit values; or as a mean value and a standard deviation, such that the window comprises a predetermined number of standard deviations about the mean. By making the stored windows narrow, an increased discrimination is provided between true money items and frauds. However, if the windows are made too narrow, the rejection rate of true money items increases, disadvantageously. The width of the windows is thus selected as a compromise between these two factors. Attempts to defraud coin or banknote acceptors typically involve the manufacture of facsimile coins or banknotes, which cause the acceptor to produce parameter signals which lie within the stored acceptance windows. Hitherto, coin acceptors have been provided with relatively wide and narrow window widths so that the operator can manually select the wide window width for normal operation and the narrow window width if frauds are being presented for validation. An example is described in Japanese unexamined patent application no Hei 2-197985.
A number of different approaches have been proposed to vary the window width dynamically to improve discrimination between true and false coins. In U.S. Pat. No. 5,355,989, a coin acceptor is described which switches automatically from a first normal acceptance window for a true coin, to a second narrower window when a coin parameter signal produced by testing a coin falls in a region of the normal window for the true coin corresponding to a low acceptance probability region for the coin concerned. A group of fraudulent coins may all have similar characteristics and they may cause the acceptor to produce parameter signals which lie within the normal window, but the parameter signals consistently have a value which is not centred on the high probability peak region of the window associated with the true coin and instead are centred on the lower probability tail regions of the bell curve distribution within the normal window. When the parameter signal falls within this low probability region, the second narrower window is then used for the next tested coin. If the next coin has a parameter falling in the narrower window it is a true coin, but if not, it is a fraud that should be rejected. This approach seeks to prevent frauds carried out by the use of coins of a particular low value denomination, from a foreign currency set, with characteristics that correspond but are not exactly the same as a high value coin of the currency set that the acceptor is designed to accept. It will be understood that the foreign denomination coins exhibit their own generally Gaussian distribution of parameter signals, and if the low probability or tail region of this distribution partially overlaps a corresponding region of the distribution for the true coin that the acceptor is designed to accept, then the low value foreign coins will sometimes be accepted as true coins.
Another approach is described in EP-A-0480736, in which the acceptance window is based on the value of a coin parameter for previous acceptable coins, as long as the previous coin parameter values do not deviate significantly from one another. This enables the coin acceptor to self-tune the window to take account of changes in operating parameters such as temperature and other long term drifts. A danger with this approach is that the coin acceptor can be taught to modify its window so as to accept frauds by using fraudulent coins similar to true coins. To overcome this problem, a so-called near miss area is defined and if a coin parameter signal from a coin under test falls in this area, this indicates the risk of a fraud and the window is shifted away from the area to prevent the window position being influenced by the potential fraud. However, the position of the near miss area is critical in order to avoid falsely detecting true items as a fraud attack. To this end the near miss area must be a reasonable distance outside of the true coin population (particularly if the error in positioning the centre of the window is taken into account). This creates a gap were a sufficiently close fraud attempt can still trigger a window shift before it is spotted in the near miss area. It may also be possible to utilise slightly modified true coins or even a different fraud on the other side of the window to train the window towards the original fraud attempt. The method described in EP-A-0480736 is therefore only of use for relatively poor quality frauds and a more stringent system is needed to counter a stronger fraud attack.